From Pathology to Policy: EMA’s first AI Qualification and What the US Should Do Next

Artificial intelligence (AI) continues to exhibit unprecedented growth and expansion as use cases reach not only important daily decisions, but also professional fields such as finance and medicine. Alongside this growth, legislative bodies must establish a sound legal framework to mitigate any significant risks to health, safety, and fundamental rights.

Specifically in the pharmaceutical field, AI now plays a larger role across the entire drug development lifecycle, from initial discovery to post-market surveillance.[i] AI has shown early success in drug candidate identification and molecular compound predictions, indicating potential for faster drug development and cost reduction.[ii] On March 20, 2025, the European Medicines Agency (EMA) took a defining step by issuing a public Qualification Opinion (QO) for AIM-MASH, an AI-assisted pathology tool used to help assess liver biopsy histology in clinical trials for new Metabolic Dysfunction-Associated Steatohepatitis (MASH) treatments.[iii] In other words, the QO recognized an AI-assisted workflow as regulatory-grade evidence when the qualified tool is locked and used exactly as specified.[iv] However, a qualified pathologist must continue to review the data, and any major change to the setup may require re-qualification.[v]

In liver disease clinical trials, human experts score liver-biopsy slides to determine eligibility and effectiveness.[vi] Specifically, in MASH drug development, these readings are especially challenging because experts can look at the same biopsy and come to completely different conclusions—variance that inflates trial size, blurs treatment effects and stalls innovation.[vii] Against that backdrop, the QO is significant because it determined that AIM-MASH can assess  “disease activity with less variability than the current standard used in clinical trials, which relies on a consensus by three independent pathologists.”[viii] The QO is neither a device clearance nor a marketing authorization, but a public acknowledgement that AI-assisted outputs will be accepted as evidence.[ix] It also sets extensive upfront requirements, mainly traceable documentation, explicit assessment of data, and measures to address potential bias or discrimination.[x] Importantly, because the decision is public, other sponsors may license the tool to reproduce the same locked version and single-reader workflow for a potentially enhanced MASH drug.[xi]

Although the EMA has taken a significant step in recognizing and regulating AI in drug development, the QO is issued “without prejudice” to other regulatory regimes such as the EU AI Act (the Act).[xii] However, these rules work together instead of being in conflict or duplicates. This is because EMA’s guidance reflects the Act’s overarching rules but is specifically tailored to the pharmaceutical sector.[xiii]

The FDA has taken a different regulatory approach but shares the same commitment to ensure safe and effective AI implementation in drug development.[xiv] In January 2025, the FDA issued a draft guidance that sets out a risk-based credibility framework for the use of AI in drug development.[xv] Specifically, a case-by-case method is adopted where sponsors must show that their model, data, and governance plan are credible for the stated purpose.[xvi] Unlike EMA’s more formal and rigorous up-front qualification, the FDA adopts a flexible structure that emphasizes transparency, adaptability, and ongoing monitoring.[xvii] In practice, the FDA seeks to develop frameworks through individualized guidance within each application.[xviii] Although this informal guidance allows for flexibility and avoids legal challenges, some contend that it reduces accountability and regulatory certainty because FDA has discretion to deviate from it published guidance.[xix]

To promote shared precedents, the FDA should keep its risk-based framework outlined in the draft guidance but also add a public, tool-specific qualification track. When a model version and experiment setup are approved, the FDA should publish that determination so other sponsors can faithfully replicate it. This would preserve the FDA’s emphasis for transparency and life-cycle rigor even while further accelerating drug innovation. Additionally, even within a case-by-case framework, the FDA should require baseline documentation comparable to EMA’s expectation. This would reduce intercontinental regulatory friction and would help eliminate barriers for pharmaceutical companies looking to get drug approval in both the US and EU.

David Chloe is a staff member of Fordham International Law Journal Volume XLIX. 

[i] See Ashfaq Ur Rehman et al., Role of Artificial Intelligence in Revolutionizing Drug Discovery, 5 Fundamental Rsch. 1273, 1283 (2024) (discussing the use of AI throughout the entire drug development process).

[ii] See id. at 1279.

[iii] See Press Release, Eur. Meds. Agency, EMA Qualifies First Artificial Intelligence Tool to Diagnose Inflammatory Liver Disease (MASH) in Biopsy Samples (Mar. 20, 2025), https://www.ema.europa.eu/en/news/ema-qualifies-first-artificial-intelligence-tool-diagnose-inflammatory-liver-disease-mash-biopsy-samples.

[iv] See id. (explaining that a qualified tool is locked when “the machine learning model cannot be modified or replaced.”)

[v] Eur. Meds. Agency, Qualification Opinion for Artificial Intelligence-Based Measurement of Non-alcoholic Steatohepatitis Histology in Liver Biopsies to Determine Disease Activity in NASH/MASH Clinical Trials 3 (2025), https://www.ema.europa.eu/system/files/documents/other/qualification-opinion-artificial-intelligence-based-measurement-non-alcoholic-steatohepat_en_1.pdf (“The AIM-NASH tool is proposed as a supplement to pathologist review and is not a substitute. The tool is always intended to be used in conjunction with the assessment of a qualified liver pathologist.”).

[vi] See Vlad Ratziu et al., Artificial Intelligence-Assisted Digital Pathology for Nonalcoholic Steatohepatitis: Current Status and Future Directions, 80 J. Hepatology 335, 335, 338 (2024), https://www.sciencedirect.com/science/article/pii/S0168827823051826.

[vii] Hanna Pulaski et al., Clinical Validation of an AI-Based Pathology Tool for Scoring of Metabolic Dysfunction-Associated Steatohepatitis, 31 Nature Med. 315, 316, 319 (2025), https://www.nature.com/articles/s41591-024-03301-2.

[viii] See Press Release, Eur. Meds. Agency, supra note 3.

[ix] Id.

[x] Gabriela Lenarczyk et al., The Future of AI Regulation in Drug Development: A Comparative Analysis, 12 J. L. & Biosciences, at 5 (forthcoming 2025), https://pmc.ncbi.nlm.nih.gov/articles/PMC12598624/pdf/lsaf028.pdf.

[xi] See supra note 5, at 3 (“Consequently, the method can be used for all MASH clinical trials in which the histologic evaluation of liver tissue is used as part of the inclusion criteria, and/or efficacy evaluation.”).

[xii] See id., at 2.

[xiii] See supra note 10.

[xiv] See U.S. FDA, Artificial Intelligence for Drug Development, https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development (last visited Nov. 10, 2025) (explaining FDA’s commitment to ensure safe and effective drug development while carefully implementing AI to facilitate innovation).

[xv] See U.S. FDA, Draft Guidance: Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products 5 (2025).

[xvi] See supra note 10 (explaining that FDA’s case-by-case method requires ongoing dialogue and adaptation).

[xvii] Marco Righi, AI in Pharma: How the FDA and EMA Are Shaping the Future of Drug Development, Life Sci. Sol. (Aug. 4, 2025), https://rpngroup.com/insights/ai-in-pharma-how-the-fda-and-ema-are-shaping-the-future-of-drug-development/.

[xviii] See supra note 10.

[xix] Id.

This is a student blog post and in no way represents the views of the Fordham International Law Journal.